Science & Tech · AI · 3 hrs ago
Microsoft says its new AI model used less computing power than OpenAI's
Microsoft says its MAI-Base-2 model outperformed a comparable OpenAI model in private tests, according to AI chief Mustafa Suleyman.
He told Microsoft employees the model used one-quarter of the computing power OpenAI used to train its model.
Training is the early stage when an AI model learns patterns from large amounts of information, and it requires substantial computing resources.
Suleyman said Microsoft’s comparisons used technical information it has about OpenAI’s models, and he also claimed MAI-Base-2 used less computing power than a DeepSeek model.
Microsoft and OpenAI are major partners, and Microsoft has invested billions of dollars in the AI company, offers its models through Azure and has used them in products such as Copilot.
Their relationship has become strained, while Microsoft has increasingly worked to develop its own models.
Microsoft announced MAI-Base-2 on Thursday, and Suleyman has been assigned to focus on building models in the company’s superintelligence unit.
Microsoft announced its MAI-Base-2 model and said it used one-quarter of the computing power used to train an OpenAI model.
Microsoft AI CEO Mustafa Suleyman said the new model was stronger than the model that powers GPT-5.6.
Suleyman said Microsoft’s model performed better in tests of understanding and predicting patterns in information.
He said the model used 12 times less computation than DeepSeek, without specifying which DeepSeek model.
Microsoft and OpenAI remain major partners, although their relationship has frayed as Microsoft develops its own models.
- Who
- Microsoft, with claims made by Microsoft AI CEO Mustafa Suleyman.
- What
- Microsoft announced MAI-Base-2 and said it used less computing power to train than comparable models from OpenAI and DeepSeek.
- When
- Thursday, October 15, 2026.
- Where
- Not stated.
- Why
- Not stated.
This story does not have two clearly opposing sides.
Microsoft announced its MAI-Base-2 model.
Suleyman told Microsoft employees at a town hall that the model used one-quarter of the computing power used to train an OpenAI model.
- Model
- MAI-Base-2
- Microsoft AI CEO
- Mustafa Suleyman
- OpenAI model cited
- GPT-5.6
- Computing power claim
- One-quarter of the amount OpenAI used to train its model
- DeepSeek comparison
- 12 times less computation, according to Suleyman
What people are saying
We now have a stronger pre-trained base model than the model that currently powers GPT 5.6 Sol
And in fact, we used one-quarter of the compute that OpenAI used to train their model.
Privately and confidentially, we've also compared it to the OpenAI models